RFC: extend signmessage/verifymessage to P2SH multisig
6 messages, 14 Apr 2013. Started by Peter Todd.
Peter Todd (3), Alan Reiner (2), Luke Dashjr
Currently signmessage/verifymessage only supports messages signed by a single key. We should extend that to messages signed by n-of-m keys, or from the users point of view, P2SH multisig addresses.
rpc.cpp:signmessage() returns the output of SignCompact(). That in turn starts with a header byte marking the signs of the various keys to allow for key recovery. The header byte can be one of 0x1B, 0x1C, 0x1D or 0x1E
For multisig signmessage signatures this is extended:
<01> <varint n> <varint m> <sig or key> {<sig or key>, ...}
Each signature or key can be one of the following forms:
sig: <1B/1C/1D/1E> <32 byte r> <32 byte s>
compress key: <02/03> <32 byte x>
uncompressed key: <04> <32 byte x> <32 byte y>
Note that we have to provide all pubkeys, even if they aren't used for a given signature, to allow the P2SH address to be reconstructed.
Decoding/encoding is a bit code-intensive due to the all the cases, but on the other hand this format keeps the size down to an absolute minimum. Alternatively I could use length bytes.
The format is backwards compatible in the sense that older versions will fail safely on new signatures, even ones that have been truncated. Similarly new signatures are easily distinguished from old, and going forward if we for some reason need yet another signature format the leading byte can be incremented.
Signing incomplete signatures on messages can be handled by converting pubkeys to signatures. Similarly the RPC signmessage command can be extended with an optional "existing signature" option.
If we're going to extend/expand message signing, can we please add a proper ASCII-armored format for it? Really, anything that encodes the signed message next to the signature, so that there's no ambiguities about what was signed. You can keep the "bare signatures" as an option for backwards compatiblity, but offer this as the primary one.
What we really want is to have the user copy an ASCII-armored block of text into the client (or we could have a URI-extension for this), and the app pops up with a window that says "The following message has a valid signature from address 1XKjf32kJbf...: <message>".
I know people argue they'd like to get away from raw addresses and copy-and-paste. But it'll be a while before that happens, and there's a lot of demand for Armory to become compatible with Bitcoin-Qt signing. People are obviously using it.
-Alan
On 04/14/2013 01:09 AM, Peter Todd wrote:
Quoted text (50 lines)
> Currently signmessage/verifymessage only supports messages signed by a
> single key. We should extend that to messages signed by n-of-m keys, or
> from the users point of view, P2SH multisig addresses.
>
> rpc.cpp:signmessage() returns the output of SignCompact(). That in turn
> starts with a header byte marking the signs of the various keys to allow
> for key recovery. The header byte can be one of 0x1B, 0x1C, 0x1D or 0x1E
>
>
> For multisig signmessage signatures this is extended:
>
> <01> <varint n> <varint m> <sig or key> {<sig or key>, ...}
>
> Each signature or key can be one of the following forms:
>
> sig: <1B/1C/1D/1E> <32 byte r> <32 byte s>
> compress key: <02/03> <32 byte x>
> uncompressed key: <04> <32 byte x> <32 byte y>
>
> Note that we have to provide all pubkeys, even if they aren't used for a
> given signature, to allow the P2SH address to be reconstructed.
>
> Decoding/encoding is a bit code-intensive due to the all the cases, but
> on the other hand this format keeps the size down to an absolute
> minimum. Alternatively I could use length bytes.
>
> The format is backwards compatible in the sense that older versions will
> fail safely on new signatures, even ones that have been truncated.
> Similarly new signatures are easily distinguished from old, and going
> forward if we for some reason need yet another signature format the
> leading byte can be incremented.
>
> Signing incomplete signatures on messages can be handled by converting
> pubkeys to signatures. Similarly the RPC signmessage command can be
> extended with an optional "existing signature" option.
>
>
>
> ------------------------------------------------------------------------------
> Precog is a next-generation analytics platform capable of advanced
> analytics on semi-structured data. The platform includes APIs for building
> apps and a phenomenal toolset for data science. Developers can use
> our toolset for easy data analysis & visualization. Get a free account!
> http://www2.precog.com/precogplatform/slashdotnewsletter
>
>
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists•sourceforge•net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-developmentOn Sunday, April 14, 2013 5:09:58 AM Peter Todd wrote:
Quoted text (3 lines)
> Currently signmessage/verifymessage only supports messages signed by a > single key. We should extend that to messages signed by n-of-m keys, or > from the users point of view, P2SH multisig addresses.
I think it would be wise to figure out HD wallet changes before trying to extend message signing. For privacy/safety, it would be a good idea to avoid signing with the same private key twice under any circumstances, so it might make sense to create a new address format the represent a chain of keys instead of one key or combination of keys.
Luke
On Sun, Apr 14, 2013 at 01:21:21AM -0400, Alan Reiner wrote:
Quoted text (10 lines)
> If we're going to extend/expand message signing, can we please add a > proper ASCII-armored format for it? Really, anything that encodes the > signed message next to the signature, so that there's no ambiguities > about what was signed. You can keep the "bare signatures" as an option > for backwards compatiblity, but offer this as the primary one. > > What we really want is to have the user copy an ASCII-armored block of > text into the client (or we could have a URI-extension for this), and > the app pops up with a window that says "The following message has a > valid signature from address 1XKjf32kJbf...: <message>".
I already looked into ASCII-armoring PGP style for a different project. (timestamping) Basically you'd want to follow RFC2440, section 7 "Cleartext signature framework":
-----BEGIN BTC SIGNED MESSAGE-----
Hello world! -----BEGIN BTC SIGNATURE----- IKBeCbxXHvD1TJh8ZlMySo26w5z6YZQD1xqKgbhsvlhEgcFD+kvKx4LzUz1yxg/8 IdYdBnzez77VDq3odHrVftg= -----END BTC SIGNATURE-----
Pretty simple really and doesn't need to be done prior to other signmessage changes. There may be an issue passing \r's through the RPC interface; the RFC specifies CRLF line endings.
On Sun, Apr 14, 2013 at 05:26:37AM +0000, Luke-Jr wrote:
Quoted text (10 lines)
> On Sunday, April 14, 2013 5:09:58 AM Peter Todd wrote: > > Currently signmessage/verifymessage only supports messages signed by a > > single key. We should extend that to messages signed by n-of-m keys, or > > from the users point of view, P2SH multisig addresses. > > I think it would be wise to figure out HD wallet changes before trying to > extend message signing. For privacy/safety, it would be a good idea to avoid > signing with the same private key twice under any circumstances, so it might > make sense to create a new address format the represent a chain of keys > instead of one key or combination of keys.
Sure, which is why I have the header byte so that when we do come up with a chain of keys thing, that in turn can get it's own magic number allocated.
FWIW I have an application now where a multisig signmessage would be useful.
On 04/14/2013 02:25 AM, Peter Todd wrote:
Quoted text (28 lines)
> On Sun, Apr 14, 2013 at 01:21:21AM -0400, Alan Reiner wrote: >> If we're going to extend/expand message signing, can we please add a >> proper ASCII-armored format for it? Really, anything that encodes the >> signed message next to the signature, so that there's no ambiguities >> about what was signed. You can keep the "bare signatures" as an option >> for backwards compatiblity, but offer this as the primary one. >> >> What we really want is to have the user copy an ASCII-armored block of >> text into the client (or we could have a URI-extension for this), and >> the app pops up with a window that says "The following message has a >> valid signature from address 1XKjf32kJbf...: <message>". > > I already looked into ASCII-armoring PGP style for a different project. > (timestamping) Basically you'd want to follow RFC2440, section 7 > "Cleartext signature framework": > > -----BEGIN BTC SIGNED MESSAGE----- > > Hello world! > -----BEGIN BTC SIGNATURE----- > IKBeCbxXHvD1TJh8ZlMySo26w5z6YZQD1xqKgbhsvlhEgcFD+kvKx4LzUz1yxg/8 > IdYdBnzez77VDq3odHrVftg= > -----END BTC SIGNATURE----- > > Pretty simple really and doesn't need to be done prior to other > signmessage changes. There may be an issue passing \r's through the RPC > interface; the RFC specifies CRLF line endings. >
Perfect. That was the spec I was looking for but too lazy to find it. I guess I'll lead by example and just do it in Armory. I'll let users pressure the other devs to follow :)
- -Alan